Evolution and Evidence in MANET Routing: From Classical Protocols to Graph Learning

This survey examines the evolution and empirical evidence of mobile ad hoc network (MANET) routing from classical protocols to learning- and graph-based approaches. Thirty-one literature workstreams are synthesized through a common framework that characterizes routing by decision locus, information scope, temporal horizon, mechanism, objective, and resource requirement, while evidence is classified by verification, comparability, and maturity. The synthesis is complemented by a controlled ns-3.47 benchmark of AODV, DSDV, the evaluated ns-3 DSR implementation, and OLSRv1 across four pre-specified scenarios and 20 predeclared seeds, yielding 320 validated confirmatory runs. AODV and DSR show no reproducible directional separation under baseline, mobility-stress, or load-stress conditions. Under scale stress, however, the paired AODV-minus-DSR packet-delivery-ratio difference is 0.1625 (bootstrap 95% CI [0.0280, 0.2950]; Holm-adjusted Wilcoxon p = 0.0437). Seed-level rankings vary across conditions, and zero-delivery incidence increases under scale stress. The evidence therefore supports condition-indexed comparison rather than universal protocol ordering. Across routing generations, the principal unresolved obligations concern reproducibility, robustness, out-of-distribution generalization, resource accounting, and simulation-to-real correspondence.